UNIVERSITI PUTRA MALAYSIA

STATISTICAL PROCESS CONTROL ANALYSIS

IBARHIM SHEIKH ABDUL KADIR HAGI

FSAS 1999 35

STATISTICAL PROCESS CONTROL ANALYSIS

IBARHIM SHEIKH ABDUL KADIR HAGI

MASTER OF SCIENCE

UNIVERSITI PUTRA MALAYSIA

1999

STATISTICAL PROCESS CONTROL ANALYSIS

By

IBRAHIM SHEIKH ABDUL KADIR HAGI

Thesis Submitted in Fulfillment of the Requirements for the Degree of Master of Science in the Faculty of

Science and Environmental Studies Universiti Putra Malaysia

March 1999

TO MY PARENTS

ACKNOWLEDGEMENTS

Firstly, praise be to God, for giving me the strength and patience to complete this

work. I would like to thank my supervisor Dr. Habshah Binti Midi for her excellent

supervision, invaluable guidance, helpful discussions and continuous encouragement,

without which this work could not be finished. My thanks also to the member of my

supervisor committee, Associate Professor Dr. Muhamad Idreess Ahmad and Dr

Kassim bin Haron for their invaluable discussions, comments, and help. Similar

thanks must go to Dr Mahendran Shitan for his help in S-plus programming

Language.

111

TABLE OF CONTENTS

Page

ACKNOWLEDGEMENTS.. ................................................... ............................ .. 111 LIST OF TABLES................................................................................................. Vll LIST OF FIGURES........................................................................................... ..... Vlli ABSTRACT............................................................................. . . . . . . . ........ ......... IX ABSTRAK........................... .................................................................................. Xl

CHAPTER

I

II

INTRODUCTION .............................................................................. .. .. Statement of the Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Some Key Words and Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Specification and Tolerance Limits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. Assignable Causes or Special Causes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Chance Causes or Common Causes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

The Objectives of the Study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Organisation of the Thesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

CONSTRUCTION OF THE CONTROL CHARTS .......... . . . . . . . . . . .

Introduction .............................................................................. .

1 7 9 9 1 0 1 1 1 1 1 2

1 4 14

The x Control Chart Based on Sample Mean . . . . ............ . . . . . . . . . . . . . '" . 1 7 The R Control Chart ...... , . . . . . . . . . . . . , . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22 The x Control Chart Based on Sample Median ..... . . . . . . . . . . . . . ... , . . . . . . . . . 22 Results and Discussion. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26 Conclusion 32

III PROCESS CAP ABILITY INDICES UNDER NON-NORMALITY ROBUSTNESS ............ , . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ' " 33 Introduction. . . . . . . ................. ................. ................................................ . . . . . 33

The Process Capability Ratio or the Cp Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . 34 The Cpk Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . 36

Clement Method . .... . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 8 John-Kot Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40

Results and Discussion 45 Conclusion 45

IV BOOTSTRAP CONFIDENCE INTERVAL ESTIMATES OF PROCESS CAPABILITY INDICES....... ............. ... ................ 47 Introduction .... . .. . . . ..... . . . . . ...................... . ... . . . . . . . . . . . . . . . . . . . . . . . " 4 7 The Method of Bootstrap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49 The Percentile Confidence Interval . . . . . .. . . . . . . . . . .. . . . . . . . . . . . . . . .. ... . . . . . . . . 50 The Bias Corrected and Accelerated Method ( Bca) . . . . . . .... . .. ... . . . .... . 51

iv

v

PERPUSTAKAAN The Bootstrap Algorithm for Estimating Bca . . 'UNIVKail11 ' PtYIltA: MALAYSlA53 The Bootstrap Algorithm for Estimating Percentile . . . . . . . . . . . . . . . . . . . . . . . . . 55

A Simulation Investigation into Bootstrap Confidence

Interval Properties for Cpk • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • 55 Results and Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . , 60

Conclusions an Recommendation . . . . . . . . . . . . , .............. , .... '" . .. . .. ... . .. 61

CONCLUSION AND SUGGESTION FOR FURTHER RESEARCH ............... , . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . .. . . . . . 63

BIBLIOGRAPHY .. 65 APPENDICES

VITA ........ ..

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 69

Appendix A Standardized values of Lp,Up, and M . ................. . . A-I Tablela. Standardized Tails of Pearson .............. . A-2 Table Ib Standardized Tails of Pearson Curves ....

69 70 72

A-3 Table 2. Standardized Median of Pearson Curves... 74 Appendix B Splus Programs for the Bootstrap Confidence

Intervals for Normal and Skewed Processes............ 76 B-1 Percentile, Bias Corrected and Accelerated, and

Standard Bootstrap Confidence Intervals for Normal Process ............................................. 77

B-2 Percentile, Bias Corrected and Accelerated, and Standard Bootstrap Confidence Intervals for

Skewed Process (Chi-square with 4.5 degree of freedom).................................................... 80

B-3 Percentile, Bias Corrected and Accelerated, and Standard Bootstrap Confidence Intervals for Skewed Process (t distribution with 8 degree of freedom)................................................... 83

86

v

Table

1

2

3

4

5

6

7

8

LIST OF TABLES

The Upper and Lower Control Limits of a Control Charts .. . . . . . . .. . . . . 25

Clement Procedure for Normal Process . .. . . . . . . .. . . . . . . . . . . . . . . . . .. . . ... . .. . 43

Results of Cp and Cpk obtained from Clement, Traditional and John

Kot Methods For Normal Process . . . . . . . . . . . . . . . . .. . .. .. . . . . . .. .. . . . . . . . . 43

Clement Procedure for Non-normal Process . . .... . .... . . . . .. . . . . . . . . . . . . . . . . 44

Results of Cp and Cpk obtained from Clement, Traditional and John

Kot Methods for Non-normal Process .. . . .. .. . . . . . . . . . . . ' " .. . . . ... . . . , . . 44

Bootstrap 90% Confidence Intervals for a Normal Process for(n=25 ,

n=50) . . . .. . . . .. . . . ... . . . . . . . . . . . . . . . . . . . . .. . . . .. . . .. . . . . . . . . . . . .. . .. . . ... . . .. . . . . . . 57

Bootstrap 90% Confidence Intervals for Non-normal Process (Chi­

square with 4.5) degree of freedom for(n=25 , n=50) . . ... . . . . . . .. . . .. . . . . . . 58

Bootstrap 90% Confidence Intervals for Skewed process (distribution

of t with 8 degree of freedom) for (n=25 , n=50) . ..... . ... . '" ... . . . . . ... . . 59

vi

LIST OF FIGURES

FIGURES Page

1 Standard Mean Control Chart without Contamination . . . . . . . . . . . . . . . . . . , . .. . .. . .. . . 28

2 Robust Standard Mean Control Chart without Contamination. . . . . . . . . . . . . . . ... . 28

3 Standard Mean Control Chart with 30% Contamination . . . . . . . . . . . . . . . . '" . , . . . , . . 29

4 Robust Standard Mean Control Chart with 30% Contamination. .. . . . . .. . . . . . . . . . 29

5 Standard R Control Chart with 30% Contamination. .. . ...... . . . . . .. . . . . . . . . . . . . .. 30

6 Standard R Control Chart with 30% Contamination . . . . . . . . . . . . , . ... .. .... .. . ... .. 30

7 Median Control Chart in Conjunction with a Chart for Sample Range without

Contamination . . . , ............................................................. " . .. . .. . . 31

8 Median Control Chart in Conjunction with a Chart for Sample Range with 30%

Contamination . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

vii

31

Abstract of thesis submitted to the Senate ofUniversiti Putra Malaysia in fulfillment of the requirements for the degree of Master of Science

STATISTICAL PROCESS CONTROL ANALYSIS

By

IBRAHIM SHEIKH ABDULKADIR HAG I

March 1999

Chairman: Habshah Bt Midi, Ph.D.

Faculty: Science and Environmental Studies

This thesis is concerned with the investigation of the two key aspects of

statistical process control. The first aspect is maintaining a stable process so that the

pattern of variation of process out-put is not changing. In order to maintain a stable

process, the study includes an examination of the state of control of the process. A

traditional variable control charts, x and R charts and also the x control chart based

on sample median and median control chart in conjunction with a chart for sample

range were used for both normal and non-normal process. The second aspect depicts

the process capability. Assuming that the processes have reached the state of

statistical control, capability measurements were proceeded in this study for both

normal and non-normal processes.

A simulation studies are carried out to compare the performance of the

traditional and the robust control chart. Likewise, the classical capability index is

compared to two robust capability index. The results of the study indicate that the

traditional and the robust control chart are equally good when no contamination in the

viii

data. However, the later performs better than the former in the presence of outliers in

the data. S imilarly, the traditional process capability index are almost as good as the

robust capability index as proposed by Clement ( 1 989) and John Kot ( 1993) in a well

behaved data. Nevertheless , the robust capabili ty index were found to be better

compared to the traditional index when contamination occurs in the data.

The study also carried out an investigation of properties of the three types of

bootstrap confidence interval for estimating the process capability index ( Cpk )'

namely the standard, percentile and bias corrected and accelerated for two processes

(normal and skewed). The average lengths displayed a consistent pattern where the

longest intervals were the standard intervals, and with the shortest intervals being the

percentile and bias corrected and accelerated intervals for both normal and highly

skewed processes. The results of the study seem to be consistent for sample size

n = 25 to n = 50 .

ix

Abstrak tesis yang dikemukakan kepada Senat Universiti Putra Malaysia bagi

memenuhi keperluan untuk ijazah Master Sains.

ANALISA PROSES KAWALAN BERSTATISTIK

Oleh

IBRAHIM SHEIKH ABDUL KADIR HAGI

March 1999

P engerusi: Habshah Bt Midi, Ph.D.

Fakulti: Sains dan Pengajian Alam Sekitar

Tesis ini berkenaan dengan penyelidikan dua aspek utama dalam proses

kawalan berstatistik . Aspek yang pertama ialah penyelenggaraan proses yang stabil

supaya corak variasi proses output tidak berubah. Bagi menentukan proses yang

stabil, kajian ini meliputi pemeriksaan keadaan bagi proses yang terkawal. Carta

pembolehubah kawalan tradisi , .x dan carta R dan juga carta kawalan .x berasaskan

median sampel dan carta kawalan median yang berkaitan dengan carta bagi sampel

julat telah digunakan bagi proses normal dan tak normal. Aspek yang kedua

menerangkan proses keupayaan. Dengan menganggapkan bahawa proses telah sampai

ke tahap kawalan berstatistik, ukuran keupayaan diteruskan dalam kaj ian ini bagi

kedua dua proses normal dan tak normal.

Kajian simulasi telah dijalankan untuk membandingkan carta kawalan tradisi

dan carta kawalan teguh. Index keupayaan klasik juga dibandingkan dengan dua

index keupayaan teguh. Keputusan kajian menunjukkan bahawa prestasi carta

kawalan teguh adalah sarna baiknya dengan carta kawalan tradisi apabila data dalarn

x

keadaan 'bersih'. Walabagaimana pun, prestasi carta kawalan teguh adalah lebih bail:

daripada carta kawalan tradisi apabila wujud data terpencil di dalam data. Begitu juga

dengan keputusan perbandingan diantara index keupayaan proses tradisi dan index

keupayaan proses teguh yang dicadangkan oleh Clement (1 989) dan John Kot (1 993).

Kedua-dua kaedah menunjukkan prestasi yang agak sarna baiknya apabila data dalam

keadaan bersih. Namun begitu, index keupayaan teguh didapati lebih baik jika

dibandingkan dengan index keupayaan tradisi apabila terdapat data 'kotor' di dalam

data.

Kajian ini juga dijalankan bagi memeriksa sifat-sifat bagi tiga jenis selang

keyakinan bootstrap bagi menganggarkan index keupayaan proses (Cpk), iaitu selang

keyakinan piawai , persentil dan pincang dibetulkan dan dipecut (Bca) bagi dua proses

(nonnal dan pencung). Purata panjang seiang menunjukkan corak yang konsisten

dengan selang yang paling panjang ialah selang keyakinan piawai dan kedua-dua

selang keyakinan persentil dan BCa adalah lebih pendek bagi kedua dua proses

nonnal dan proses pencung. Keputusan kajian ini kelihatan konsisten bagi setiap saiz

sampel n=25 dan n=50.

xi

CHAPTER I

INTRODUCTION

Statistical methods provide a very effective means for the development of

new technology and quality control in manufacturing process. Many leading

manufacturers have been striving for an active use of statistical methods, and some

of them spend more than 100 hours annually in in-company education on this

subject. While knowledge of statistical methods is becoming part of the normal

fixture of an engineer, the fact that one knows statistical methods does not

immediately lead to the ability to use it. The ability to treat maters from the

statistical viewpoint is more important that the individual methods. In addition, it is

not easy to chose among the different statistical techniques that can be used in

solving a given problem.

One of the most important statistical tools for the quality control in

manufacturing process is Statistical Process Control (SPC); the goal of statistical

process control is to maintain a stable process where the pattern of variation of the

process output is not changing. Statistical process control may be addressed in

terms of three key aspects:

2

a) Process control: maintaining the process on target with respect to centering and

spread.

b) Process capability: Determining the inherent spread of a controlled process for

establishing realistic specifications.

c) Process change: Implementing process modification as a part of process

improvements and troubleshooting.

Process capability indices (an aspect ofSPC) are one of the focuses of this study.

The use and abuse of process capability indices have been the subject of

considerable controversy in the last few years, their widespread and often uncritical

use may almost inadvertently have led to some improvement in quality, but also,

almost certainly, have been the cause of many unjust decision, which might have

been avoided by better knowledge of their properties.

The fundamental task of process capability indices are to determine whether

a manufacturing production process is capable of producing items within

specification limits and that is customer requirements (Kane, 1986; Rado, 1989;

Montgomery, 1985).

Process capability indices are used widely throughout industries, to give a

relatively quick indication of process capability in a format that is easy to compute

and understand.

3

Among the widely used process capability indices are Cp and Cpk' Each of

these indices are indicative of process ability to satisfy customer requirements (i.e.

specification limits). Process capability indices such as Cp and Cpk are assumed to

have the properties of normality. The C p index is used to measure the potential of a

process to satisfy a customer requirements (i.e. specification limits). The

manufacturer in the form of the specification limits usually sets the allowable

process spread, and the actual process spread is based on the distribution (usually

estimated) of the product obtained from manufacturing process. The allowable

upper and lower specification limits as denoted as (USL and LSL) respectively, the

actual process spread usually contained in the natural tolerance limits which are

considered to be six times the true standard deviation of the process from

manufacturing product. The C pk index is developed to indicate how process

confirms to two-sided specification limits. This C pk index is used to measure

process performance. The manufacturing engineer recommended value of C p and

Cpk equals to 1 .33 is the minimum value that should be observed for the acceptable

process capability.

One of the advantages of the process capability indices are that it provides

an easily understandable aggregate measure of goodness of the process

performance. The ability to meet specifications is the criterion used for measuring

the attractiveness of the process. The capability described above are non-

4

dimensional, which makes them more versatile and appealing because they do not

depend on the specific process parameter units (Kane, 1 986).

The determination of the capability of a process should begin only after the

process has been brought to a state of statistical control. A process is said to be in

statistical control when the only source of variation in the system is a result of

chance causes. The use of control chart is an important step, which must be taken in

early stages in an SPC program to eliminate assignable causes, reduce variability,

and stabilize process performance.

In this study, a traditional variable control charts commonly known as mean

(x) and range (R) control charts were used to examine the state of control of the

process. This will be the first stage of SPC procedure. The variable control charts

such as x and R charts were chosen because it allows studying a process regardless

of its specifications. The x and R charts are also allow to employ both

instantaneous variable (short-term process capability), and variability across the

(long-term process capability). It is particularly helpful if the data for a process

capability study are collected in two to three different time periods, such as

different shift and different day's . The variable control charts are important in the

quality program of many industries, their ability to identify process improvement

opportunities. Inherent in the construction of control charts for variables is the

assumption that the process under examination is normally distributed with

independence and identical observations. Continuous follow process often has

5

outocorrelated observations, which violate the independence assumption and render

standard control charts for variables unreliable. In order to rectify this problem a

more robust control charts are needed such as mean ( x ) control chart based on

sample median and median control chart in conjunction with a chart for sample

range R for individual observations. The final solution is not easily effected by

outocorrelation.

Assuming that the processes have reached a state of statistical control, the

capability measurements can then proceeded. The C p and C pk indices have been

calculated to determine whether a manufacturing production process is capable of

producing items within specification limits. The manufacturing engineers

recommend that the values of Cp and Cpk which equals to 1 .33 is the minimum

value that should be observed for acceptance process capability. But the question is

often arises is whether it can meet the tolerance. As mentioned earlier, it is sensible

to estimate the ability of a process to meet specification limits only when it is in a

state of statistical control. In that state, the process has no assignable causes, so

exhibited process variability is a reflection of what the process can achieve. A

process should first be analyzed to verifY that it is in control before its capability is

estimated. An assumption that always is made is that, the process output (which is,

the distribution of the quality characteristic under consideration) is normal. The

assumption of normality can be validated by means of empirical plots of histogram,

normal probability plots, or statistical test for goodness of fit, such as Chi-square

tests or the Kolmogorov- Smirnov Test ( Cochran, 1 952; Duncun, 1986; Massey,

6

1952; Nelson, 1986; Shapiro, 1980). A study by Gunter (1989), have criticized the

validity of traditional method in using the capability measurements specially when

the underlying distribution is not normal, such approach ignores the fact that the

Cp and Cpk are random variables. However, when capability measures are used, it

is worth noting that some processes do not follow normally distributions, perhaps,

due to the presence of autocorrelations or outliers in the data. The discussion of

non-normality falls into two main parts. The first and easier of the two, is

investigation of the properties of process capability indices and their estimators

when the distribution of process has specific non-normal forms. The second and

more difficult, is development of the methods to cuter for the non-normality and

consideration for the use of new process capability indices specially designed to be

robust (i.e . , not too sensitive) to non-normality. Only recently statisticians have

provided a new methods which are considered to be an alternative to traditional

method when underlying distribution is not normal.

Many studies to improve process capability indices when underlying

distribution is not normal have been done. The papers are too numerous to be

reviewed on non-normal process capability indices (Kane 1986; Gunter 1 99 1 ; Pean

et aI., 1 992; English and Bates 1 99 1, Kocherlakota 1 992; Subramaniam 1 966, 1 968;

Karl pearson 1 983; Clement 1 989;John-Kot 1 992; Munecchilka 1 986, 1 992;

Johnson and Kot 1 970; Chan 1 988 ; Dovich 1 99 1 ,a, b; McCoy 191; Chan, L.K.,

Cheng, S.W., and Spring, F.A. 1 988; Chan and Zhang 1990; Johnson, Kortz and

Pearn 1 992; Hall, P. and Martin 1 988, 1 989; Owen and Borrego 1 990; Marucucci

7

and Beazaly 1 988; Mirbella 199 1 ; Spring 1 99 1 a; Subbaih and Toom 199 1 ;

Kushler and Hurly 1 992; Johnson 1 993; Balistskaye and Zoatuhina 1 988; Hall

1 992; Gunter 1 989; Franklin and Wassermann 1991 ; Price 1 992; Goh 1 994;

Mooney and Duvall 1 993).

However, the effects of non-normality on properties of process capability

indices have not been major research items until quite recently. The number of

different capability indices have increased and so lead to confusion among

practitioners, as such were unable to provide an adequate and clear explanation of

the meaning of various indices, and more importantly when the underlying

distribution is not normal. Literature review indicated that some aspects of these

problems have not been solved completely, even the problems solved so far have

different approaches. It is important to recognize non-normality to avoid gross

errors, not only in capability measurements but also for correct prescription of

control chart techniques and other aspects of statistical analysis. Therefore, we

should incorporate robust method in our study so as to correct the problems of

outliers and outocorrelations if it exist in the data.

Statement of the Problem

Many manufacturing production processes have problem in determination

of process capability whether, it can produce items within specification limits, and

this is because capability indices such as Cp and Cpk are assumed to have

properties of normality. Gunter ( 1 989) pointed out, many processes are normal with

skewed distribution, however, and approximate technique based on traditional

8

method may perform badly and can be misleading, especially when underlying

distribution is not normal. Literature showed that some researchers have dealt with

non-normal process capability indices but their approaches in dealing with non­

normal process are different, and non-recommendable.

This research is concerned with the three key aspects of SPC as mentioned

earlier. To examine the state of control of the process, variable control charts such

as x and R charts were used as a traditional method, where robust standard mean

control chart and median in conjunction with a chart for sample range were used as

robust method. The study also focused on the estimation of capability

measurements for normal and non-normal processes. Traditional method used

normal process, whereas non-normal process used robust method, which was

discussed by (Clement, 1989; John-Kot, 1 992).

The study also focuses on the estimation of confidence interval limits for

Cpt by using non-parametric bootstrap method. The percentile, bias corrected and

accelerated confidence intervals for C pk index were calculated from bootstrap

method. Such intervals represent a major step toward the correct understanding and

interpretation of C pk index. Bootstrap confidence interval for estimating C pk does

not depend on the usual assumption of normality and in fact can be calculated

regardless of whatever the underlying process distribution was.

9

Some Key Words and Definitions

For the sake of completeness, a brief review of some important concepts

that were used frequently in this study are given below. Some important key words

include Specification limits and tolerance limits, assignable causes, and common

causes.

Specification and Tolerance Limits

Specification and tolerance limits are often used interchangeable and are

defined by the ANSIIASQC standard A l ( 1 987) as the limits that define the

conformance boundaries for an individual unit of a manufacturing or servIce

operation. The standard suggests that the tolerance limits are generally preferred in

evaluating the manufacturing or service requirements, whereas specification limits

are more appropriate for categorizing materials, products, or services in terms of

their stated requirements. In general, tolerances are subsets of specifications.

Usually, tolerances pertain to all requirements.

Tolerance limits may be two-sided (with upper and lower limits) or one­

sided with either upper or lower limits. A lower tolerance limit defines the lower

conformance boundary for an individual unit of a manufacturing or service

operation; an upper specification limit is determined by the needs of the customer.

What the customer wants in the product service is analyzed by means of market

10

research and incorporated through product or servIce design. These limits are

placed on a product characteristic by designers and engineers for an individual unit

in order to ensure an adequate functioning of the product . Some part of this study

uses the term's specification limits and tolerance limits interchangeably because of

the ANSIIASQC standard A l (1987) makes no distinction between them.

Assignable Causes or Special Causes

Variability caused by special causes or assignable cause that is not

inherent in the process. That is, it is not part of the process as designed and does not

effect all items. Special causes can be use of a wrong tool, an improper raw

material, or an operator error. If an observation falls outside the control limits or

non-normal pattern is exhibited, special causes are assumed to exist, and the

process is said to be out of control . One objective of control chart is to detect the

presence of special causes as soon as possible to allow appropriate corrective

action. Once special causes are eliminated through remedial actions, the process is

again brought to state of statistical control . Deming ( 1 982) believed that 1 5% of all

problems are due to special causes. Actions on the part of both management and

workers will reduce the occurrence of such causes.

11

Chance Causes or Common Causes

Variability due to common causes is inherent to a process. It exists as long

as the process is not changed and is referred to, as the natural variation in a process.

It is inherent part of the process design and effects all items. This variation in the

effect of many small causes and cannot be totally eliminated. When this variation is

consisiting, we have what is known as a stable system of common causes. A

process operating under a stable system of common causes is said to be III

statistical control. Examples include inherent variation of incoming raw materials

from a qualified vendor, a lack of adequate supervision skills, the vibration of

machines, and fluctuations in working condition. Management alone is responsible

for common causes. Deming (1982) believed that about 85% of all problems are

due to common causes and hence can be solved only by action on the part of

management. In a control chart, if quality characteristic values are within the

control limits and no non-random pattern is visible, it is assumed that a system of

common causes exists and the process is in state of statistical control.

The Objectives of the Study

The main objectives of the study are

1. To ensure the stability of the manufacturing process by using robust median

control chart and median control chart in conjunction with chart for a sample

range R.

12

2. To employ a robust method which may improve capability of the process when

underlying distribution is not normal. Our focus of the study is limited to the

statistical process control for variables.

Other aspects that will also be considered in this study are

i) To compare the performance of traditional and robust control charts

ii) To compare the performance of traditional and robust process capability

index

iii) To adopt the non-parametric bootstrap method on the Cpk index.

Organisation of the Thesis

This thesis is comprised of five chapters.

Chapter I, introduces the objectives of the research, and depicts the need to

use robust estimation procedure instead of traditional methods. Chapter II describes

the construction of control charts by using both traditional and robust methods. The

traditional control chart was based on sample mean whereas the robust control

charts were based on sample median. A complete discussion on process capability

measurement was exhibited, in Chapter III after assuming that the processes have

reached a state of statistical control. The discussions address the process capability

index in both situations where the process having a normal and non-normal

distribution. In this chapter the classical C pk is compared to the robust methods

such as Clement and John Kot techniques. In Chapter IV the parametric bootstrap

method for evaluating percentile, bias corrected and accelerated confidence